Generating Optimal Discrete Analogue of the Generalized Pareto Distribution under Bayesian Inference with Applications

• Published in: Symmetry (Basel) Vol. 14; no. 7; p. 1457
• Main Authors: Haj Ahmad, Hanan, Almetwally, Ehab M.
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.07.2022
• Subjects: Asymmetry; Bayesian analysis; Bayesian estimation; Coronaviruses; COVID-19; Discretization; discretization methods; Failure times; Goodness of fit; Methods; Monte Carlo Markov chain; Parameter estimation; prior distribution; Random variables; simulation analysis; Statistical inference; symmetric and asymmetric loss functions
• ISSN: 2073-8994; 2073-8994
• DOI: 10.3390/sym14071457

This paper studies three discretization methods to formulate discrete analogues of the well-known continuous generalized Pareto distribution. The generalized Pareto distribution provides a wide variety of probability spaces, which support threshold exceedances, and hence, it is suitable for modeling many failure time issues. Bayesian inference is applied to estimate the discrete models with different symmetric and asymmetric loss functions. The symmetric loss function being used is the squared error loss function, while the two asymmetric loss functions are the linear exponential and general entropy loss functions. A detailed simulation analysis was performed to compare the performance of the Bayesian estimation using the proposed loss functions. In addition, the applicability of the optimal discrete generalized Pareto distribution was compared with other discrete distributions. The comparison was based on different goodness-of-fit criteria. The results of the study reveal that the discretized generalized Pareto distribution is quite an attractive alternative to other discrete competitive distributions.